Influence of High Level Features of HVS on Performance of FSIM

Abstract

In this paper the influence of information about high level features of Human Visual System (HVS) on objective quality assessment is studied. This was done by extending the existing full-reference objective image quality metric – FSIM – where the different importance of certain areas of image is considered using Phase Congruency (PC) algorithm. Here, the estimation of Region of Interest (ROI) based on this algorithm is complemented by Fixation Density Maps (FDM) containing the information about high level features of HVS. Use of another low level features based algorithm (Phase Spectrum of Fourier Transform) was also considered and compared to the PC algorithm. The performance was evaluated qualitatively on images reconstructed according to ROI and quantitatively on images from LIVE database. The correlation between subjective and objective tests was calculated using Pearson’s Correlation Coefficient and Spearman’s Rank Order Coefficient. The statistical significance of the difference between correlation coefficients was assessed by Fisher r-to-z transformation. The performance of the metric was also compared to other state-of-the-art image quality metrics (SSIM, MS-SSIM, and FSIM).

Keywords

• Human visual system
• image quality assessment
• high level features
• FSIM